Sindhu Pawar

AI Researcher

Bengaluru, Karnataka, India3 yrs 2 mos experience

Key Highlights

Expert in developing voice agents and ASR systems.
Proven track record in optimizing LLM training processes.
Skilled in building multilingual voice solutions.

Stackforce AI infers this person is a SaaS specialist with expertise in voice technology and machine learning.

Contact

Skills

Core Skills

Automatic Speech RecognitionNatural Language Processing (nlp)Machine LearningLarge Language Models (llm)Text-to-speech SynthesisData ScienceVoice Cloning

Other Skills

Computer VisionDPOData AnalysisDeep LearningDeepspeedPython (Programming Language)RLHFVoice Agent

Experience

Level ai

Machine Learning Engineer

Jun 2024 – Sep 2025 · 1 yr 3 mos · Bengaluru, Karnataka, India · Remote

Built an end to end voice agent from scratch using opensource LLM and OpenAI Streaming TTS
Leading end-to-end deployment of real-time ASR using Triton Inference Server and conformer ASR model optimizing inference speed by 10x for 300 concurrent clients and achieved 4x higher throughput
Developed a BERT-based punctuation model for partial ASR outputs with 95% accuracy

Voice AgentAutomatic speech recognitionNatural Language Processing (NLP)Python (Programming Language)Machine LearningDeep Learning

Krutrim

Founding Data Scientist

Jun 2022 – May 2024 · 1 yr 11 mos · Bengaluru, Karnataka, India · On-site

Worked on DPO(Direct Preference Optimization) in LLMs for solving neutrality, safety and biasness in responses. Successfully trained a 7B parameter model using DPO and LoRA technique
Optimized LLM training using ZeRO Stage 2/3, reducing training time significantly upto 4x on distributed machine
Improved voice-cloning in Text-to-speech(TTS) models for indian languages and built a voice assistant model for a chatbot
Developed a real-time multilingual live stream converter (5 Indian languages) with synchronized lip-sync with a 10-minute lag along with voice cloning (dubbing of video and audio)
Voice Acivity Detection (VAD) Implemented a 1d cnn for VAD with an optimal decoding strategy
Ranker Algorithm (maps) Implemented an xgboost ranker algorithm for the search recommendations in Ola Maps
Built a Real Time Voice Command Detection bot to identify commands to operate the scooter os

DPOLarge Language Models (LLM)Text-to-Speech SynthesisVoice CloningData ScienceMachine Learning

Indian institute of technology, bombay

Research Intern

Jan 2022 – May 2022 · 4 mos · Mumbai, Maharashtra, India · Hybrid

* Worked on knowledge distillation for the fine grain classification of bird images with RESET-50 (student) and RESNET-158 (teacher) based on the difference in the distribution of heatmaps and divergence between logits

Data ScienceMachine Learning